Related Literature

This chapter is related to studies analyzing behavior after apologies for actually experienced and economically relevant offenses (Ohtsubo and Watanabe (2009), Ho (2007), Skarlicki et al. (2004), Schweitzer et al. (2006), Bottom et al. (2002), Abeler et al.

(2010)). Our contribution is the distinction between intentionally committed offenses and offenses with ambiguous intentionality. We are the first to analyze how an apology affects punishment. Our study is also the first one that analyzes offenders’ motives for sending an apology.

In psychology there is a large body of evidence that a harmdoer who sends an apology is much more likely to be forgiven than a non-apologizer. Most of these studies

can be grouped into three categories. In the first category psychologists present vignettes describing situations in which an offender did or did not apologize.

Participants then make judgments about the offender (See for example Ohbuchi and Sato (1994), Weiner et al. (1991), Girard et al. (2002), Ohtsubo and Watanabe (2009), Wada (1998), Scher and Darley (1997)). The second category includes studies where participants have to remember past self-experienced situations. They are told to recall whether the offender did or did not apologize and to give explanations of how they felt in this particular situation and whether they accepted the apology. (See for example Exline et al. (2007), McCullough et al. (1997), McCullough et al. (1998), Schmitt et al.

(2004).) The third category uses deceptive role-play with actual offenses (Ohbuchi et al.

(1989), Struthers et al. (2008)). All three categories document that apologies have a mitigating effect on anger and increase forgiveness.

Apologies are most effective when they are sufficiently long and come across as sincere (Darby and Schlenker (1989), Shapiro (1991), Skarlicki et al. (2004)), include an expression of responsibility (Scher and Darley (1997), Struthers et al. (2008)), an expression of remorse (Gold and Weiner (2000), Tavuchis (1991), Darby and Schlenker (1989)) and explanations in the form of excuses rather than justifications (Shaw et al.

(2003)). It is not clear yet how offers of compensation affect forgiveness. On the one hand, several studies show that compensation payments can increase forgiveness (Bottom et al. (2002), Schmitt et al. (2004), Scher and Darley (1997), Witvliet et al.

(2002), Zechmeister and Romero (2002)). On the other hand, Abeler et al. (2010) find that customers who receive an apology instead of monetary compensation forgive significantly more often. The authors conduct a field experiment with three different treatments: An apology treatment, where the customer receives an email including an apology and a high and a low compensation treatment. The authors argue that getting paid money could reduce the intrinsic motivation of customers to forgive (as in Gneezy and Rustichini (2000)) and that an apology might trigger a heuristic to forgive that is hard to overcome rationally. In opposition to this reasoning are the results by Struthers et al. (2008), Skarlicki et al. (2004) and Bennett and Earwaker (1994). They show that an apology does not trigger a heuristic to forgive but that its mitigating effect on punishment crucially depends on the characteristics of the offense. The literature thereby clearly distinguishes between responsibility and intentionality of the harm. In a vignette study Bennett and Earwaker (1994) analyze whether the offender’s

responsibility affects the acceptance of an apology. They find that the higher the responsibility for the harm, the lower the acceptance of the apology. In Struthers et al.

(2008) forgiveness was less likely following an apology when offenders intentionally committed an offense. Skarlicki et al. (2004) present very similar results. In their study receivers of unfair offers in an ultimatum game accept these offers less often after an apology than after no message was sent. Since in this case an unfair offer is always made intentionally, these results show that after intentionally committed harm an apology can backfire and even increase punishment. Those who apologize for an intentionally committed harm may be perceived as self-interested, untrustworthy, and as having an ulterior motive (Fein (1996), Schul et al. (2004)). This might lead to lower acceptance rates. Ohbuchi and Sato (1994) conduct experiments with children and find this effect also with fifth graders. The children accepted a harmdoer’s apology only when they believed that the harm was committed unintentionally. Interestingly, second graders were not sensitive to the harmdoer’s intent.

Ohtsubo and Watanabe (2009) and Ho (2007) introduce theoretical models predicting that receivers of an apology are sensitive to the cost involved in the apology.

Experimental evidence is ambiguous. Ohtsubo and Watanabe (2009) find their hypotheses confirmed. In their experiment participants in the costly apology condition abstained from sending a complaint message to their offender after an apology. In the repeated trust game by Ho (2007), costs do not affect the event of forgiveness, which is measured by the amount entrusted in the following period.

An apology’s effect on reputation has been analyzed in Schweitzer et al. (2006) and Bottom et al. (2002). Schweitzer et al. (2006) use a trust game in order to test how apologies can repair trust after an offense within a repeated interaction. Schweitzer et al. (2006) find that an apology alone does not facilitate trust recovery. The apology has to come along with a promise for future trustworthy behavior. Bottom et al. (2002) conduct a prisoner’s dilemma and find that apologies indicating good intentions for the future have a positive effect on trust recovery in repeated interactions.

Apologies also differ with respect to culture and gender. Asians apologize more than Americans (Takaku et al. (2001)) and women apologize more than men (Tavuchis (1991), page 127). Frantz and Bennigson (2005) find that apologies expressed at a later stage of a conflict are more effective than earlier ones, and that this effect is mediated by feeling heard and understood.

In law there is a broad literature on apologies, too. Here the main question of interest is whether an apology is a possible mechanism to avoid a law suit. For overviews see Cohen (2002) or White (2009).

2.3 Experimental Design and Procedure

Our basic design is similar to Ho (2007) who uses a trust game with an apology option at the end. We use a sequential prisoner’s dilemma with apology option and punishment. Additionally we manipulate the mechanism how to cooperate. For cooperation, subjects have to correctly answer a question. Defection results, when they answer the question incorrectly. This feature allows that defection can be the result of inability or of intentional unkindness. In detail, the sequence is the following:

1. At the beginning of the game both players receive an endowment of 60 points in order to avoid negative payoffs throughout the game.

2. Player A then receives a multiple choice question.

If he gives the correct answer, he loses 40 points and his partner player B receives 120 points. If he answers incorrectly, points do not change.

3. Both players learn about player A’s result. Next, player B receives a multiple choice question, too. If he answers it correctly, he loses 40 points and player A receives 120 points. If he answers incorrectly, points do not change. Therefore, if both players answer their questions correctly, they receive 80 points each. However, player B maximizes his payoff by giving a wrong answer after a correct answer by player A.

4. The players learn whether the answers were correct and player B can send an individual message to player A. We are interested in whether offenders use the message to apologize for their harming.

5. Next, player A can deduct points from player B. One deducted point costs player A 0.2 points. Punishment is restricted such that it cannot yield negative payoffs.

The questions were easy but not trivial. For example, we asked for the capital of Japan, giving Tokyo, Osaka, Yokohama and Kyoto as possible answers. By ensuring that not all participants could perfectly solve the questions, we induced uncertainty of

intentions. Not giving the right answer to the question can be chosen intentionally¹² or due to inability. We provide a measure for the question’s difficulty and an opportunity for the players to form an individual view on the difficulty of their partners’ questions.

For that purpose the players receive their partners’ multiple choice questions at the same time. If they answer this solo question correctly, they receive 5 points. We call the fraction of players who are able to solve their partner’s multiple choice question the solvability benchmark. On average participants could solve 85% of the questions.¹³

We run five treatments. The first treatment is the control treatment (baseline) as explained above. In the second treatment no punishment no punishment is possible. We use this treatment to analyze whether offenders still apologize when they do not fear punishment. Third, in the treatment costly writing a message costs 5 points. The results of this treatment will show whether the decision to apologize depends on costs and whether costly apologies are more credible. In the fourth treatment no apology option, player B cannot send messages. We thereby control how people cooperate and punish deviators when no apology is possible. The fifth treatment no quiz varies the potential assignment of intentions. In the quiz games a wrong answer can be due to intentional harm or due to inability. In no quiz, an offense is always caused intentionally. In this treatment we use the same parameters as before. However, participants do not have to answer questions. They just decide whether they want to give up 40 points in order to cede 120 points to their partner. Table 2.1 gives an overview of the treatments.

Baseline No

punishment Costly No apology

option No quiz

The whole procedure was common knowledge. We conducted 12 sessions in the time from June to November 2009. All sessions were conducted at the LakeLab

12 In a laboratory experiment Utikal and Fischbacher (2009) analyze how people attribute intent to helpful and harmful actions. They find that when the harming agent is economically strong, people perceive the offense as intentional. Therefore we define one possible reason for giving a wrong answer as intent.

13 We discuss the implications in the results section.

(TWI/University of Konstanz) with a total number of 356 participants. Before the experiment started, subjects were randomly assigned to their role as player A or B.

Players kept their role throughout the game in order to avoid learning the content of others’ messages. The experiment lasted 10 rounds. We used a perfect stranger matching in order to avoid repetition effects and to keep players A from receiving identical messages. Participants received the income of all periods. One point translated into 0.01 euros. The experiment took about 60 minutes, average income of participants was 10.93 euros (14.87$) plus a show-up fee of 2 euros (2.72$). The games were programmed with z-Tree (Fischbacher (2007)). We recruited participants using the online recruiting system ORSEE (Greiner (2004)). Each subject sat at a randomly assigned PC terminal and was given a copy of instructions.¹⁴ A set of control questions was provided to ensure the understanding of the game. The experiment did not start until all subjects had answered all questions correctly. We ensured that no subject participated more than once in our experiment. We rule out spillover effects across and within sessions by giving every player a different question.